create_thread validation is not synchronized with app-server model/list

Summary

The Codex app's create_thread tool can reject or hang for an external model that the bundled Codex app-server returns from model/list.

Environment

• Codex desktop: 26.730.61639
• Bundled core: codex-cli 0.147.0-alpha.1.2
• External catalog model: vibeproxy/deepseek-v4-flash

A fresh app-server model/list returns the namespaced model and its supported reasoning efforts. The exposed create_thread tool description still omits it from the advertised model list.

Reproduction

1. Configure an external catalog containing vibeproxy/deepseek-v4-flash.
2. Confirm that app-server model/list returns it.
3. Call create_thread(model="deepseek-v4-flash", thinking="low", ...).

• Actual: reasoning-effort validation rejects the combination.

1. Call create_thread(model="deepseek-v4-flash", ...).

• Actual: create_thread could not validate model "deepseek-v4-flash".

1. Call with the correct catalog ID, model="vibeproxy/deepseek-v4-flash".

• In this environment the call no longer returns the immediate validation error, but hangs and produces no discoverable rollout.

Expected behavior

The tool schema/validator should be derived from the active app-server model/list response, including the model's advertised reasoning efforts. If a returned model cannot be used for thread creation, fail quickly with the downstream capability reason instead of a stale allowlist or a hung request.